manmal

24 ms·

631.
by manmal 2y ago
I‘ve configured Little Snitch to only allow it access to huggingface. I think for updates I need to reset LS to „ask for each connection“ or sthg like that.
632.
by manmal 2y ago
Using a 7B model on a M2 Max also isn’t quite the most impressive way to locally run an LLM. Why not use QwQ-32 and let it give some commercial non-reasoning models a run for their money?
633.
by manmal 2y ago
I wish there were a variation that can destructure more ergonomically. Instead of: ``` fetch_data() |> (fn {:ok, val, _meta} -> val :error -> "default value" end).() |> String.upcase() ``` Something like this
634.
by manmal 2y ago
Just create a screenshot of the pic and upload that.
635.
by manmal 2y ago
Copilot experimental (need VSCode Insiders) has it. I‘ve thought about trying aider —-watch-files though, also works with multiple files.
636.
by manmal 2y ago
The bureaucracy of internationalizing it would likely be more expensive than the current cost.
637.
by manmal 2y ago
Remember „You are a bad user, I am a good bing“? Apple is just slower in fixing and improving things.
638.
by manmal 2y ago
You get some superficial checking by the compiler and test cases, but hallucinations that pass both are still an issue.
639.
by manmal 2y ago
Is that the four letter domain PG recently tweeted about? Congrats!
640.
by manmal 2y ago
Isn’t a problem there that a cache would be model specific, where the cached items are only valid for exactly the same weights and inference engine? I think those are both heavily iterated on.
641.
by manmal 2y ago
Isn’t that actually also in favor of building up more complexity?
642.
by manmal 2y ago
I don’t know about valuing kids - maybe they are valued enough, in a very direct sense. But, as a parent of three kids, it’s very very obvious that modern western society is not really made for bringing up children in a good way. I could li
643.
by manmal 2y ago
I wouldn’t have expected that everyone has as many friends as themselves, hence my confusion. Thanks for taking the time to respond though.
644.
by manmal 2y ago
I don’t fully understand how this is a paradox. I guess it’s based on the fact that there are „super nodes“ in human social networks, where one person has many friends, while most others don’t? Or is there anything more to it?
645.
by manmal 2y ago
We are looking into adopting bidirectional streams, and have identified gRPC as a likely ideal candidate. It provides a layer on top of the blobs (partial responses) sent by either side, and takes over the required chunking and dechunking.
646.
by manmal 2y ago
Another example is fast fashion. It’s almost impossible to get clothing in a store that will last, say, 10 years. While in earlier decades, that was pretty normal.
647.
by manmal 2y ago
> increasingly shrinking pool of people and organizations What do you mean, is the pool really shrinking, worldwide? I must have missed some statistics?
648.
by manmal 2y ago
> For example, I spent $30 for Atari video games in the 1980s and it was a lot less expensive to produce. That game would cost $90 today with inflation. And the Nintendo Switch has sold 5x as many consoles as did Atari. Likely a similar
649.
by manmal 2y ago
Personally, I‘m holding off on purchasing a new iPhone for this reason, even though my 13 Pro is getting long in the tooth as a full time iOS dev. The coming generation is rumored to have more and better memory (LPDDR5X?), and cooling (vapo
650.
by manmal 2y ago
I think prompt processing also needs all the weights.
651.
by manmal 2y ago
Won’t prompt processing need the full model though, and be quite slow on a Mac?
652.
by manmal 2y ago
For coding, I‘m finding Claude‘s responses most to the point and on-task. While many other models try to extrapolate or lecture or patronize. DeepSeek is pretty good though. Maybe it’s the high latency (probably due to prompt processing)?
653.
by manmal 2y ago
> I have noticed many more errors are detected in AI-generated Rust code than in any other language. Is that because the Rust compiler is just a very strong guardrail? Sounds like it could work well for Swift too. If only xcodebuild were
654.
by manmal 2y ago
> Fluoridated water is the reason I've never had a cavity despite growing up poor with a terrible diet, dental hygiene and care. I don’t follow the leap here. You could just be blessed with a good oral microbiome, proper calcium, D3
655.
by manmal 2y ago
In Swift, String? is an alias to Optional<String>. Optional is an enumeration with two cases: some, and none, where the former has the non-nil value attached. I‘m not sure how that ties in with subtyping though.
656.
McMaster researchers discover new class of antibiotics(brighterworld.mcmaster.ca)
2 points by manmal 2y ago|2 comments
657.
by manmal 2y ago
With some LEDs, and especially the warm whites, I get this feeling of „artificiality“, for lack of a better word. Like the surroundings are somehow fabricated. It’s an interesting idea to mix and match those with other light sources, I‘m go
658.
by manmal 2y ago
May I ask why you don’t get a used 3090 with 24GB VRAM?
659.
by manmal 2y ago
Looks like the A4000 has low memory bandwidth (50% of a 4090?) which is the limiting factor for inference usually. But they are efficient - if you can get them for cheap, probably a good entry setup? If you like running models that need a l
660.
by manmal 2y ago
Is that a problem? According to this, the GPUs don’t communicate that much once the weights are loaded: https://github.com/turboderp/exllama/discussions/16#discussi... > So at FP16 precision that's a

More ›